Main

In your final repo, there should be an R markdown file that organizes all computational steps for evaluating your proposed Facial Expression Recognition framework.

This file is currently a template for running evaluation experiments. You should update it according to your codes but following precisely the same structure.

if(!require("EBImage")){
  install.packages("BiocManager")
  BiocManager::install("EBImage")
}

package 㤼㸱EBImage㤼㸲 was built under R version 4.0.3

if(!require("R.matlab")){
  install.packages("R.matlab")
}

package 㤼㸱R.matlab㤼㸲 was built under R version 4.0.4

if(!require("readxl")){
  install.packages("readxl")
}

package 㤼㸱readxl㤼㸲 was built under R version 4.0.3

if(!require("dplyr")){
  install.packages("dplyr")
}

package 㤼㸱dplyr㤼㸲 was built under R version 4.0.3

if(!require("readxl")){
  install.packages("readxl")
}

if(!require("ggplot2")){
  install.packages("ggplot2")
}

package 㤼㸱ggplot2㤼㸲 was built under R version 4.0.3

if(!require("caret")){
  install.packages("caret")
}

package 㤼㸱caret㤼㸲 was built under R version 4.0.4

if(!require("glmnet")){
  install.packages("glmnet")
}

package 㤼㸱glmnet㤼㸲 was built under R version 4.0.3

if(!require("WeightedROC")){
  install.packages("WeightedROC")
}

package 㤼㸱WeightedROC㤼㸲 was built under R version 4.0.4

if(!require("keras")){
  install.packages("keras")
}

there is no package called 㤼㸱keras㤼㸲WARNING: Rtools is required to build R packages but is not currently installed. Please download and install the appropriate version of Rtools before proceeding:

https://cran.rstudio.com/bin/windows/Rtools/
trying URL 'https://cran.rstudio.com/bin/windows/contrib/4.0/config_0.3.1.zip'
Content type 'application/zip' length 81597 bytes (79 KB)
downloaded 79 KB

trying URL 'https://cran.rstudio.com/bin/windows/contrib/4.0/tensorflow_2.2.0.zip'
Content type 'application/zip' length 144784 bytes (141 KB)
downloaded 141 KB

trying URL 'https://cran.rstudio.com/bin/windows/contrib/4.0/tfruns_1.5.0.zip'
Content type 'application/zip' length 1486258 bytes (1.4 MB)
downloaded 1.4 MB

trying URL 'https://cran.rstudio.com/bin/windows/contrib/4.0/zeallot_0.1.0.zip'
Content type 'application/zip' length 62029 bytes (60 KB)
downloaded 60 KB

trying URL 'https://cran.rstudio.com/bin/windows/contrib/4.0/keras_2.3.0.0.zip'
Content type 'application/zip' length 2487967 bytes (2.4 MB)
downloaded 2.4 MB

package ‘config’ successfully unpacked and MD5 sums checked
package ‘tensorflow’ successfully unpacked and MD5 sums checked
package ‘tfruns’ successfully unpacked and MD5 sums checked
package ‘zeallot’ successfully unpacked and MD5 sums checked
package ‘keras’ successfully unpacked and MD5 sums checked

The downloaded binary packages are in
    C:\Users\evesu\AppData\Local\Temp\RtmpyMTkrd\downloaded_packages

if(!require("tensorflow")){
  install.packages("tensorflow")
}

package 㤼㸱tensorflow㤼㸲 was built under R version 4.0.4

library(R.matlab)
library(readxl)
library(dplyr)
library(EBImage)
library(ggplot2)
library(caret)
library(glmnet)
library(WeightedROC)
library(tensorflow)
library(keras)

package 㤼㸱keras㤼㸲 was built under R version 4.0.4

Step 0 set work directories

set.seed(2020)
# setwd("~/Project3-FacialEmotionRecognition/doc")
# here replace it with your own path or manually set it in RStudio to where this rmd file is located. 
# use relative path for reproducibility

Provide directories for training images. Training images and Training fiducial points will be in different subfolders.

train_dir <- "../data/train_set/" # This will be modified for different data sets.
train_image_dir <- paste(train_dir, "images/", sep="")
train_pt_dir <- paste(train_dir,  "points/", sep="")
train_label_path <- paste(train_dir, "label.csv", sep="") 

Step 1: set up controls for evaluation experiments.

In this chunk, we have a set of controls for the evaluation experiments.

• (T/F) cross-validation on the training set
• (T/F) reweighting the samples for training set
• (number) K, the number of CV folds
• (T/F) process features for training set
• (T/F) run evaluation on an independent test set
• (T/F) process features for test set

run.cv <- TRUE # run cross-validation on the training set
sample.reweight <- TRUE # run sample reweighting in model training
K <- 5  # number of CV folds
run.feature.train <- TRUE # process features for training set
run.test <- TRUE # run evaluation on an independent test set
run.feature.test <- TRUE # process features for test set

Using cross-validation or independent test set evaluation, we compare the performance of models with different specifications. In this Starter Code, we tune parameter lambda (the amount of shrinkage) for logistic regression with LASSO penalty.

lmbd = c(1e-3, 5e-3, 1e-2, 5e-2, 1e-1)
model_labels = paste("LASSO Penalty with lambda =", lmbd)

Step 2: import data and train-test split

#train-test split
info <- read.csv(train_label_path)
n <- nrow(info)
n_train <- round(n*(4/5), 0)
train_idx <- sample(info$Index, n_train, replace = F)
test_idx <- setdiff(info$Index, train_idx)

If you choose to extract features from images, such as using Gabor filter, R memory will exhaust all images are read together. The solution is to repeat reading a smaller batch(e.g 100) and process them.

n_files <- length(list.files(train_image_dir))

image_list <- list()
for(i in 1:100){
   image_list[[i]] <- readImage(paste0(train_image_dir, sprintf("%04d", i), ".jpg"))
}

Fiducial points are stored in matlab format. In this step, we read them and store them in a list.

#function to read fiducial points
#input: index
#output: matrix of fiducial points corresponding to the index
readMat.matrix <- function(index){
     return(round(readMat(paste0(train_pt_dir, sprintf("%04d", index), ".mat"))[[1]],0))
}

#load fiducial points
fiducial_pt_list <- lapply(1:n_files, readMat.matrix)
save(fiducial_pt_list, file="../output/fiducial_pt_list.RData")

Step 3: construct features and responses

• The follow plots show how pairwise distance between fiducial points can work as feature for facial emotion recognition.

  • In the first column, 78 fiducials points of each emotion are marked in order.
  • In the second column distributions of vertical distance between right pupil(1) and right brow peak(21) are shown in histograms. For example, the distance of an angry face tends to be shorter than that of a surprised face.
  • The third column is the distributions of vertical distances between right mouth corner(50) and the midpoint of the upper lip(52). For example, the distance of an happy face tends to be shorter than that of a sad face.

Figure1

feature.R should be the wrapper for all your feature engineering functions and options. The function feature( ) should have options that correspond to different scenarios for your project and produces an R object that contains features and responses that are required by all the models you are going to evaluate later.

• feature.R
• Input: list of images or fiducial point
• Output: an RData file that contains extracted features and corresponding responses

source("../lib/feature.R")
tm_feature_train <- NA
if(run.feature.train){
  tm_feature_train <- system.time(dat_train <- feature(fiducial_pt_list, train_idx))
  save(dat_train, file="../output/feature_train.RData")
}else{
  load(file="../output/feature_train.RData")
}

tm_feature_test <- NA
if(run.feature.test){
  tm_feature_test <- system.time(dat_test <- feature(fiducial_pt_list, test_idx))
  save(dat_test, file="../output/feature_test.RData")
}else{
  load(file="../output/feature_test.RData")
}

CNN

Step 4: Train a classification model with training features and responses

Call the train model and test model from library.

train.R and test.R should be wrappers for all your model training steps and your classification/prediction steps.

• train.R
  • Input: a data frame containing features and labels and a parameter list.
  • Output:a trained model
• test.R
  • Input: the fitted classification model using training data and processed features from testing images
  • Input: an R object that contains a trained classifier.
  • Output: training model specification
• In this Starter Code, we use logistic regression with LASSO penalty to do classification.

source("../lib/train.R") 
source("../lib/test.R")

For CNN’s Averagin is better? https://datascience.stackexchange.com/questions/47797/using-cross-validation-technique-for-a-cnn-model#:~:text=Any%20time%20you%20have%20models,tendency%20toward%20overfitting%20not%20underfitting.

source("../lib/cross_validation.R")

feature_train = array(unlist(dat_train[, -6007]), dim = c(2400,152,3))
label_train = array(as.numeric(dat_train$label)) - 1 

• Train the model with the entire training set using the selected model (model parameter) via cross-validation.

# training weights

tm_train <- system.time(fit_train <- train(feature_train, label_train))

Model: "sequential_45"
_______________________________________________________
Layer (type)            Output Shape          Param #  
=======================================================
conv1d_116 (Conv1D)     (None, 150, 32)       320      
_______________________________________________________
max_pooling1d_72 (MaxPo (None, 75, 32)        0        
_______________________________________________________
conv1d_115 (Conv1D)     (None, 73, 64)        6208     
_______________________________________________________
max_pooling1d_71 (MaxPo (None, 36, 64)        0        
_______________________________________________________
conv1d_114 (Conv1D)     (None, 34, 64)        12352    
_______________________________________________________
flatten_45 (Flatten)    (None, 2176)          0        
_______________________________________________________
dense_144 (Dense)       (None, 64)            139328   
_______________________________________________________
dense_143 (Dense)       (None, 1)             65       
=======================================================
Total params: 158,273
Trainable params: 158,273
Non-trainable params: 0
_______________________________________________________
Epoch 1/10
53/53 - 0s - loss: 3.7640 - accuracy: 0.7304 - auc: 0.5108
53/53 - 1s - loss: 3.7640 - accuracy: 0.7304 - auc: 0.5108 - val_loss: 0.5971 - val_accuracy: 0.7375 - val_auc: 0.5812
Epoch 2/10
53/53 - 0s - loss: 0.4981 - accuracy: 0.8071 - auc: 0.5612
53/53 - 1s - loss: 0.4981 - accuracy: 0.8071 - auc: 0.5612 - val_loss: 0.4893 - val_accuracy: 0.7986 - val_auc: 0.6121
Epoch 3/10
53/53 - 0s - loss: 0.4737 - accuracy: 0.8107 - auc: 0.6223
53/53 - 1s - loss: 0.4737 - accuracy: 0.8107 - auc: 0.6223 - val_loss: 0.4853 - val_accuracy: 0.7986 - val_auc: 0.6225
Epoch 4/10
53/53 - 1s - loss: 0.4807 - accuracy: 0.8083 - auc: 0.6170
53/53 - 1s - loss: 0.4807 - accuracy: 0.8083 - auc: 0.6170 - val_loss: 0.5010 - val_accuracy: 0.7986 - val_auc: 0.6206
Epoch 5/10
53/53 - 0s - loss: 0.4877 - accuracy: 0.8089 - auc: 0.6037
53/53 - 1s - loss: 0.4877 - accuracy: 0.8089 - auc: 0.6037 - val_loss: 0.4873 - val_accuracy: 0.7986 - val_auc: 0.6259
Epoch 6/10
53/53 - 0s - loss: 0.4678 - accuracy: 0.8089 - auc: 0.6447
53/53 - 1s - loss: 0.4678 - accuracy: 0.8089 - auc: 0.6447 - val_loss: 0.5028 - val_accuracy: 0.7986 - val_auc: 0.6346
Epoch 7/10
53/53 - 0s - loss: 0.4595 - accuracy: 0.8095 - auc: 0.6672
53/53 - 1s - loss: 0.4595 - accuracy: 0.8095 - auc: 0.6672 - val_loss: 0.4886 - val_accuracy: 0.8056 - val_auc: 0.6405
Epoch 8/10
53/53 - 0s - loss: 0.4725 - accuracy: 0.8095 - auc: 0.6265
53/53 - 1s - loss: 0.4725 - accuracy: 0.8095 - auc: 0.6265 - val_loss: 0.4988 - val_accuracy: 0.7986 - val_auc: 0.6413
Epoch 9/10
53/53 - 0s - loss: 0.4592 - accuracy: 0.8107 - auc: 0.6661
53/53 - 1s - loss: 0.4592 - accuracy: 0.8107 - auc: 0.6661 - val_loss: 0.4778 - val_accuracy: 0.7986 - val_auc: 0.6541
Epoch 10/10
53/53 - 0s - loss: 0.4506 - accuracy: 0.8131 - auc: 0.6900
53/53 - 1s - loss: 0.4506 - accuracy: 0.8131 - auc: 0.6900 - val_loss: 0.4706 - val_accuracy: 0.8014 - val_auc: 0.6707

#save(fit_train, file="../output/fit_train.RData")

Step 5: Run test on test images

tm_test = NA

feature_test <- array(unlist(dat_test[, -6007]), dim = c(600, 152, 3))

if(run.test){
  #(file="../output/fit_train.RData")
  tm_test <- system.time(
                         {label_pred <- test(fit_train, feature_test, type = "classes");
                          prob_pred <- test(fit_train, feature_test, type = "proba")}
                         )
  
}

label_test <- as.integer(dat_test$label)
weight_test <- rep(NA, length(label_test))
for (v in unique(label_test)){
  weight_test[label_test == v] = 0.5 * length(label_test) / length(label_test[label_test == v])
}

label_test <- ifelse(label_test == 2, 1, 0)
accu <- sum(weight_test * (label_pred == label_test)) / sum(weight_test)
# prob_pred <- apply(prob_pred, 1, max)
prob_pred <- prob_pred[, 1]
tpr.fpr <- WeightedROC(prob_pred, label_test, weight_test)
auc <- WeightedAUC(tpr.fpr)

cat("The accuracy of model:", "CNN", "is", accu*100, "%.\n")

The accuracy of model: CNN is 51.7871 %.

cat("The AUC of model:", "CNN", "is", auc, ".\n")

The AUC of model: CNN is 0.720733 .

Summarize Running Time

Prediction performance matters, so does the running times for constructing features and for training the model, especially when the computation resource is limited.

cat("Time for constructing training features=", tm_feature_train[1], "s \n")

Time for constructing training features= 1.01 s 

cat("Time for constructing testing features=", tm_feature_test[1], "s \n")

Time for constructing testing features= 0.23 s 

cat("Time for training model=", tm_train[1], "s \n") 

Time for training model= 21.25 s 

cat("Time for testing model=", tm_test[1], "s \n")

Time for testing model= 0.5 s 

###Reference - Du, S., Tao, Y., & Martinez, A. M. (2014). Compound facial expressions of emotion. Proceedings of the National Academy of Sciences, 111(15), E1454-E1462.